Multivariate Hazard Assessment for Nonstationary Seasonal Flood Extremes Considering Climate Change

In recent years copulas have been widely employed in multivariate modeling of hydrological extremes. However, anthropogenic and climate changes have greatly impacted the probabilistic behavior of these extremes and have challenged the stationarity assumption of the marginal distributions of individual characteristics of the extremes as well as their dependence structure inherent in copula-based modeling. This study developed a dynamic copula-based multivariate risk analysis model (DCM) to analyze seasonal flood extremes (MWLandAMS), observed at Yichang Station in Yangtze River basin, China. The model entailed three parts: (1) Time-varying moment models, combined with log-likelihood ratio tests, were employed to explore whether the trend-caused nonstationarity existed in the marginal distributions or the dependence structure of the seasonal flood extremes; (2) a nonstationary multivariate probability distribution was developed using a dynamic marginal and copula-based model, incorporating different large-scale climate forcings (NAO, SOI, and NINO3.4), meteorological factor (rainfall and temperature) and reservoir index as covariates; and (3) flood hazard was quantified using the multivariate hazard model. The climate-related nonstationarity-based multivariate frequency model through the incorporation of climatic indices would help predict the hazard level of a certain quantile pair for the next year of the observed period.